This paper proposes a new scheme in which subphoneme units are used to reduce spectral distortion at concatenation points, thereby achieving high-quality concatenative speech synthesis. Spectral distortion is minimized by a dynamic programming technique based on a concatenation point table. The validity of the proposed method has been shown in objective and subjective evaluation tests. The objective test confirmed that the maximal spectral distortion at concatenation points was reduced by 16-44%. In the subjective test, speech synthesized using the proposed method got 60% of the preference score in comparison with that using phoneme units.
Bibliographic reference. Itoh, Yoshiharu / Hashimoto, Makoto / Higuchi, Norio (1995): "Sub-phonemic optimal path search for concatenative speech synthesis", In EUROSPEECH-1995, 577-580.